|
PureBytes Links
Trading Reference Links
|
Dale,
one of the problems you mentioned might stem from the fact that
maxbarsback of the system might be different from indicators.
The general problem I was referring to did not concern TS but
rather the logic of indicators.
If indicators are constructed like
> if price > Indicator and Low<low[2] then buy at high + 1 point stop;
> if Price < Indicator and H>H[2] then sell at l - 1 point stop;
then the buy/sell signal gets triggered by a one-off event which
might be gone on the next bar. In my opinion trade signals should be
in general constructed like flags. If the flag is on one is long
or short if the flag is off you are not in a position.
The general reason beeing: Why would one want to be in a long
position if you would NOT buy at this price level. The above code
leaves you in long or short position although the condition for
purchase or sell is not satisfied anymore.
However in my example the system does not perform anymore when one
adapts the system in such a fashion. And this points in my opinion to
so some fundamental problem with the indicator. The ideal indicator
should basically represent the propablity of a move in the predicted
direction. An indicator swinging from -100 to 100 should express the
propability of a move. The threshold at which one enters the position
represents the transaction cost which must be covered.
Gerrit
> This is the biggest problem TS 4.0 has, which everyone has been
> ignoring. The inaccurate results of comparisons in the system area.
> I've worked out some of my systems by hand and my research shows
> that a indicator may have a totally different value when referenced
> in a system - not even similar to what is plotted on the screen.
>
> It has made TS totally useless as far as system testing is
> concerned. Everything must be backtested by hand. A recent example
> of this was a system I wrote using the volatility indicator that
> Cindy Leee shared with the group a few weeks ago,.
>
> Here were the parameters.
>
> if price > Indicator and Low<low[2] then buy at high + 1 point stop;
> if Price < Indicator and H>H[2] then sell at l - 1 point stop;
>
> Even with this simple system about half of the system entries as
> shown by the price plot did not show on the system when programmed.
> After verifying a few times I got the correct signals to appear only
> to disappear again after I verified again. Never to be seen again.
> All this without changing the Code !!!!
>
> I have the TS 5.0 Preview I hope to get the password soon so I can
> test to see if it has this same problem.
>
> I've been dealing with this certain problem for 4 + years ! In two
> weeks AIQ pro with system and indicator writing abilities will be on
> the market. The language is similar to visual basic.
>
> Between TS 5.0 or AIQ we should be able to work around this finally.
>
>
> -----Original Message-----
> From: Orphelin@xxxxxxx [SMTP:Orphelin@xxxxxxx]
> Sent: Wednesday, November 04, 1998 6:48 PM
> To: Omega-list
> Subject: Re: Construction of trading signals (not indicators)
>
> Dans un courrier daté du 04/11/98 19:12:18 Heure d7iver Pari18
> Madrid, jrt@xxxxxxxxxx a écrit :
>
> > have come across some very common problem which appears when one
> > constructs trading signals.
> >
> > What has been bothering me is the following:
> >
> > Lets say I have got a two indicators which work well across a number
> > of markets and which are meant to have a predictive meaning. They
> > are implemented in the following way:
> >
> > IF (Indicator1 > threshold1) and (Indicator2 > threshold2) then buy
> > IF (Indicator1 < - threshold1) and (Indicator2 < - threshold2) then
> > sell
> >
> > Basically two conditions must be satisfied before a long position is
> > established or reversed.
> >
> > This system works well however it is wrongly constructed in my
> > opinion because if I assume that these indicators have got
> > a predictive meaning then one should not be in a position when one of
> > the conditions is NOT satsified. So from a logical point of view the
> > system should be rather:
> >
> > IF (Indicator1 > threshold1) and (Indicator2 > threshold2) then buy;
> > IF (Indicator1 <= threshold1) or (Indicator2 <= threshold2) then
> > exit long;
> > IF(Indicator1 < - threshold1) and (Indicator2 < - threshold2) then
> > sell;
> > IF(Indicator1 >= - threshold1) or (Indicator2 >= - threshold2) then
> > exit short;
> >
> > Basically the system is only in the market when both conditions
> > are satisfied and not as in the first case until a new position is
> > established.
> >
> > However this system does not perform at all profits. So what is
> > the conclusion from this exercise ?
> >
> > 1. The indicators are not valid at all.
> > 2. The indicators are only wrong for the exit signal.
> >
> > Has anyone got any suggestion how to test either of these theories ?
> >
> > Gerrit Jacobsen
> >
> >
>
> 1 and 2 are possible, but there is another one
>
> Your example is a typical one that use crisp values where you limit
> your choice to 2*2 boolean conditions, what restricts the power of
> your indicators. The complete solution could be given using fuzzy
> logic for example. No proof that the exit levels are given by the
> same values for long and short entries/ exits.
>
> You can try with our free neurofuzzy thing version 2 and your two
> indicators. I bet that it will find a better solution than any crisp
> coding. Within minutes.
>
> Sincerely,
>
> Pierre Orphelin
> www.sirtrade.com
>
|